icnc13.pdf
-
Upload
aislangabrielhernand -
Category
Documents
-
view
214 -
download
1
Transcript of icnc13.pdf
-
Cooperative Resource Allocation in OFDM-Based
Multicell Cognitive Radio Systems
Qianyu Yang, Shaowei Wang & Mengyao Ge
School of Electronic Science and Engineering
Nanjing University, Nanjing 210093, China
E-mail: [email protected], [email protected], [email protected]
AbstractIn this paper, we investigate the resource alloca-tion problem for an orthogonal frequency division multiplexing(OFDM) based multicell cognitive radio (CR) system. Secondaryusers (SUs) served by the CR system distribute randomly inmultiple cells and share radio spectrum with primary users(PUs) in a licensed system, where the interference introducedto the PUs must be kept below their tolerable thresholds. In oursystem model, SUs in different CR cells can transmit signalswith the same OFDM subchannel, so cochannel interferenceamong the SUs should also be considered. We propose an efficientalgorithm, named as multi-level waterfilling (MLWF), to allocatepower among OFDM subchannels for all CR cells by jointlyconsidering transmission power and interference constraints. TheMLWF always allocates much power to a subchannel whichgenerates less interference to the PUs. Simulation results showthat our proposed algorithm provides better performance thanother existing ones. Moreover, the complexity of the MLWF ismuch lower than other representative algorithms.
Index TermsCognitive radio, Cooperative resource alloca-tion, OFDM, Optimization
I. INTRODUCTION
Spectrum scarcity becomes a severe problem as the rapid de-
velopment of wireless service demand. However, in despite of
the looming spectrum shortage crisis, investigations show that
radio spectrum is far away from fully utilized [1]. Although
radio spectrum has already been assigned to licensed primary
users (PUs), part of them is usually unused at a certain time
or location. That is to say, spectrum holes exist in both time
and location. In order to fulfill the requirements of spectrum-
hungry applications, cognitive radio (CR) is brought up [2, 3]
and attracts more and more attention in the past ten years [4].
On the other hand, to prohibit the performance degeneration
of the PUs, the interference generated by the SUs must be
regularly controlled. Hence, the physical layer of CR systems
should be very flexible to meet these requirements.
Orthogonal frequency division multiplexing (OFDM) has
been considered as an appropriate modulation scheme for CR
systems [5], owing to its high flexibility in dynamic resource
allocation. A heuristic algorithm called Max-Min is proposed
in [6]. Simulation results show its performance is close to
the optimal, but the computational cost is relatively high. A
simple but fast efficient algorithm is implemented in [7] by
introducing a normalized index to measure the ability of a
subchannel to carry bits. It achieves performance close to the
optimal with a very low computational complexity. However,
both of the algorithms in [6] and [7] are only suitable for
single cell scenario.
For the single cell scenario, SUs are generally assumed
to transmit signals on different subchannels simultaneously.
Nevertheless, for multicell case, it is more reasonable to allow
that SUs in different cells use the same subchannel simulta-
neously [810]. In [8], a distributed algorithm is proposed
to maximize the throughput of the considered CR system
while guaranteeing the rates of nominal users. In [9], a
fully distributed subchannel selection and power allocation
algorithm is proposed by combining an unconstrained opti-
mization method with a constrained partitioning procedure.
However, interference introduced to PUs is not considered
in [8, 9]. In [10], a greedy-like heuristic method, referred to
as multicell Max-Min algorithm, is proposed to solve the
resource allocation problem in multicell CR networks. The
computational complexity of the proposed algorithm is very
high because it has to solve a set of nonlinear equations during
each iteration.
In this paper, we focus on the resource allocation in the
downlink of a multicell OFDM-based CR system and try
to maximize the overall data rate of the SUs. Firstly, we
give a greedy-like water-filling procedure to load bits for
subchannels. Each CR user greedily loads bits on a subchannel
based on its signal to interference plus noise ratio (SINR).
The MLWF applies a simple cooperative adjustment procedure
after the greedy waterfilling algorithm, which can reduce both
cochannel interference and mutual interference. By coopera-
tively moving bits from the subchannels which generates high
interference to the subchannels with lower interference, the
MLWF makes an effort to reduce the capacity loss caused
by the greedy-like algorithms. Simulation results show our
proposed algorithm has a better performance with remarkable
lower complexity compared to other existing methods.
The rest of this paper is organized as follows. Section II
describes system model and formulates the optimization prob-
lem. In III, proposed algorithm is shown in detail. Simulation
results and analysis are presented in Section IV. Conclusion is
drawn in Section V.
II. SYSTEM MODEL AND PROBLEM FORMULATION
A. System Model
Consider that multiple OFDM-based CR cells share spec-
trum with a licensed system. Each CR cell locates a base
2013 International Conference on Computing, Networking and Communications, Cognitive Computing and NetworkingSymposium
978-1-4673-5288-8/13/$31.00 2013 IEEE 724
-
2
Secondary BS1
PU2BS
Primary networkPU1
SU
Secondary BS2SU
Fig. 1. System model: Coexistence of primary and secondary cells
station (BS) and serves an SU. Fig.1 shows the system model
considered in this work. We consider the downlink of the CR
system in this work and try to maximize the sum throughput of
the SUs served by CR cells. We assume that perfect channel-
state information (CSI) is available by channel estimation at
the receiver side, and the CSI is also available at each CR
cells transmission side by feedback. Perfect CSI between
the secondary BSs and a specified PU is also available by
searching the database or sensing the channel.
The primary BS transmits to L PUs. The whole availablespectrum is divided into N subchannels with equal bandwidthB and the starting frequency is fs. There are M CR cells andeach cell is permitted to employ the N subchannels. Let N :={1, 2, . . . , N}, L := {1, 2, . . . , L}, M := {1, 2, . . . ,M}denote the set of subchannels, PUs and CR cells, respectively.
The nominal spectrum of subchannel n, n N , ranges fromfs+(n1)B to fs+nB. The PU ls nominal band is supposedto span from fPUl to f
PUl + Bl, where f
PUl is the starting
frequency and Bl is the occupied bandwidth of PU l, l L.The power spectrum density (PSD) of the OFDM subchan-
nel n used by an SU can be expressed as
SUn (f) = Ts
(
sinfTsfTs
)2
, (1)
where Ts is the symbol duration. The interference introducedto the PU l from CR cell ms BS on subchannel n with unittransmit power is
ISPm,l,n =
fPUl fs(n1
2)B+Bl
fPUl
fs(n1
2)B
gSPm,l,nSUn (f)df, (2)
where gSPm,l,n is the power gain from the CR cell ms BS tothe PU ls receiver on subchannel n.
On the other hand, assume the primary BS casts unit power
on subchannel n, the interference introduced to the SU in CRcell m on subchannel n is
IPSl,m,n =
fs+nBfPUl
1
2Bl
fs+(n1)BfPUl 1
2Bl
gPSn,mPUl (f)df, (3)
where gPSn,m is the power gain on subchannel n from primaryBS to the SU served in CR cell m. PUl (f) is the PSD of PUls signal.Since the primary network and CR network coexist in sys-
tem, there are two kinds of interference introduced to the SUs.
First, cochannel interference between CR cells arises when
multiple SU in different CR cells use the same subchannel.
That is to say, the signal of one CR cell is treated as interfer-
ence by other CR cells transmitting on the same subchannel.
On the other hand, the PUs also generate interference to the
subchannels. Let ISSm,n denote the interference to the SU of themth CR cell on the nth subchannel introduced by the otherCR cells,
ISSm,n =
j 6=m,jM
pjngSSj,m,n, (4)
where pjn is the CR cell js transmit power on subchannel nand gSSj,m,n is power gain from CR cell js BS to CR cell msuser.
The SINR of SU served in the CR cell m on subchannel ncan be defined as
Hm,n = gSSm,m,n/(
2 + IPSm,n + ISSm,n), (5)
where 2 denotes the additive white gaussian noise variance.And IPSm,n is the total interference from PUs to the subchanneln of SU served in CR cell m. From (3), it can be calculatedby
IPSm,n =L
l=1
plnIPSl,m,n, (6)
where pln is the PU ls signal power on subchannel n.The achievable transmission rate on the nth subchannel of
the mth CR cell is
rm,n = B log2 (1 + (pm,nHm,n)/) , (7)
where is the SINR gap.
B. Problem Formulation
The optimization objective is to maximize the sum capacity
of the CR cells which operate in a power-limited situation,
while keeping the interference to the PUs not exceeding spec-
ified thresholds{
IUl , l = 1, 2, . . . , L}
. Thus, the constrained
optimization problem can be formulated as follows,
maxM
m=1
Nn=1 rm,n
s.t. C1 :N
n=1 pm,n Pm, pm,n 0, m M
C2 :M
m=1
Nn=1 pm,nI
SPm,l,n I
Ul , l L.
(8)
The C1 is the power constraint of the mth CR cell. Pmis the maximum transmit power. The C2 are the interferenceconstraints, where the interference threshold of PU l is IUl ,l L.
III. THE PROPOSED ALGORITHM
The optimization problem formulated in (8) is a nonlinear
programming coupling in the constraint C2 which poses aprohibitive computational burden to the system and is gener-
ally very hard to solve. In [11], an iterative waterfilling (IWF)
algorithm yields a nash equilibrium if we regard the resource
allocation problem as a non-cooperative game. Although the
complexity of the IWF algorithm is much less than that of
the exhaustive search algorithm and converges fast in various
725
-
3
network scenarios, it has sub-optimal performance with greedy
criterion.
To address limitations of IWF, many other methods have
been proposed. In [10], a heuristic algorithm called multicell
Max-Min is proposed to solve the problem formulated in
(8). The complexity of the multicell Max-Min algorithm is
O(
RM4N)
, where R is the total number of bits allocatedto all SUs. Obviously, the computational complexity of the
algorithm is too high to be applied in practical CR systems.
We consider the problem formulated in (8) as a non-
cooperative game in our previous work [12] and propose an
iterative algorithm which decomposes (8) into m independentsub-problem. Using iterative method to update interference
level of each CR system makes the interference constrains
to be satisfied while maximizing the sum capacity of the CR
system. Moreover, in this work, we develop a more efficient
algorithm with better performance and lower complexity.
A. MLWF Optimization Model
From (8), we can see that the constraint C1 is independentfor each CR cell while the constraints C2 is not. Withoutloss of generality, suppose each SU m on subchannel n has avirtual power spectral mask (PSM) pUm,n which can keep thethe constraint C2 satisfied in (8). We rewrite the optimizationproblem as
maxM
m=1 Rm
s.t. C1 :N
n=1 pm,n Pm, m M
C2 : 0 pm,n pUm,n, m M, n N
(9)
where Rm =N
n=1 rm,n is the sum rate of cell m. To meetthe interference constraint C2 in (8), we have
M
m=1
N
n=1
pUm,nISPm,l,n I
Ul , l L (10)
We firstly solve the optimization problem (9) and then use
MLWF algorithm to ensure (10) by a power (bits) adjustment
procedure. We propose to use Lagrangian multipliers to model
the problem (9).
Define the Lagrangian
L ({pm,n}) =M
m=1
{
N
n=1rm,n m
(
N
n=1pm,n Pm
)
N
n=1m,n(pm,n p
Um,n) +
N
n=1m,npm,n
}
(11)
where {m}m=1,...,M ,and {m,n, m,n}m=1,...,M,n=1,...,N arenonnegative Lagrangian multipliers. Based on Karush-Kuhn-
Tucker (KKT) conditions, we obtain the following results.
The transmit power allocated to subchannel n for user m isgiven by
pm,n = (1/(m + m,n) 1/(m,n))+, (12)
where ()+ denotes max{, 0}, and m,n = m/Hm,n. Here,dual variables m and m,n control the waterfilling level of
SU m on subchannel n. Since m,n changes with n, each SUm has many WF levels.
Generally, to account for the transmit power and the PSM
constraints C2 in (9), following iterations based on sub-gradient search may be implemented
k+1m =km m(Pm
N
n=1
pm,n) (13)
k+1m,n =km,n m
(
pUm,n pm,n)
(14)
where 0 m 1 is a gradient search step size, and km andkm,n denote the values of m and m,n at the kth step, re-spectively. Moreover, this approach converges in theory when
the dual update stepsize m is small enough [13]. However,gradient-based search method has a very slow convergence
due to the large number of dual variables involved. To over-
come this problem, we design a multicell cooperative MLWF
algorithm which can achieve a fast and stable convergence.
B. Multicell MLWF Algorithm
We adopt a cooperative manner to allocate the power
among all subchannels. First, we use the waterfilling algorithm
to allocate the total power Pm in each cell while keepingpm,n pUm,n. If the solution already satisfies the constraints(10), we can directly obtain the power allocation for problem
(8). Otherwise, we cooperatively adjust the power (bits) to
mitigate the interference introduced by SUs to PUs.
Similar to the Levin-Campello loading [14], the MLWF al-
gorithm also maintains an incremental energy table for discrete
power (bits) adjustment procedure. With as the granularityof the discrete bits ( = 1 for integer bits or = 0.5 fora complex channel quadrature amplitude modulation signal).
Assume user ms current bit distribution is rm,n. The energyrequired to maintain the bit distribution on subchannel n ofuser m is calculated as [15]
nm(rm,n) = 2(m/Hm,n)(2rm,n 1). (15)
The incremental energy enm to load an additional bit onsubchannel n of user m is thus
enm(rm,n) = (m/Hm,n)2rm,n+1(2 1). (16)
The pseudo-code of the multicell cooperative MLWF al-
gorithm is in Table I. The third step of the algorithm is the
fixed-margin version of the greedy bit-loading procedures by
exploiting waterfilling. The fifth step performs the multicell
cooperative power (bits) adjustment procedure to reduce the
mutual interference from SUs to PUs. The key idea of this
step is to move the power (bits) for one subchannel with high
interference to PUs to the other one which generates lower
interference to PUs until the power allocation scheme satisfied
the interference constraint (10). Obviously, the main computa-
tional loads lie on the greedy power allocation procedure and
multi-level cooperative power (bits) adjustment procedure.
726
-
4
TABLE ITHE PSEUDO-CODE OF MLWF ALGORITHM
Algorithm: Multicell Cooperative MLWF Algorithm
Step 1:Initialize
1:Set rm,n 0, nm(rm,n) 0, pm,n 0, m M,
n N . Given Pm.
2:Calculate the ISPm,l,n
,IPSl,m,n
and IPSm,n respectively by (2),
(3),(4). And compute enm(rm,n + 1), m M,n N .
3:flag=1;
4:while(flag==1)
step 2:Calculate the SINR
5: Calculate the SINR by (5);
step 3:Greedy power allocation procedure
6: for m = 1 to M
7: Perform the waterfilling algorithm with Pm constraints
and obtain the pm,n and rm,n;
8: end for
step 4:Test the interference constraint
9: if the constraint (10) is met
10: flag=0, break;
11: end if
step 5: Multi-level Cooperative Power (bits) Adjustment
12: for n = 1 to N
13: g argmax(m,pm,nPm){
pm,nIm,l,n}
, m M;
14: h argmin(m,pm,nPm){
pm,nIm,l,n}
, m M;
15: if (h + eng (rg,n) e
nh(rh,n)) Ph &&
16: (g + enh(rh,n) eng (rg,n)) Pg
17: pg,n pg,n + enh(rh,n) eng (rg,n);
18: ph,n ph,n + eng (rh,n) e
nh(rh,n);
19: rg,n rg,n 1, rh,n rh,n + 1;
20: end if
21: end for
22:end while
TABLE IICOMPUTATIONAL COMPLEXITY COMPARISON
Algorithm Complexity
Exhaustive Search O(eMN )
Multicell Max-Min O(
RM4N)
Iterative algorithm O(M2N log2 N)
MLWF O(MNlog2N)
C. On the Complexity and Convergence
The MLWF greedy power allocation procedure and the
multi-level cooperative power (bits) adjustment, the dominated
parts of the whole algorithm, have the same complexity
which is equal to O(N logN) where N is the number ofsubchannels. MLWF contains a simple line-search with bi-
section or diminishing step size and thus adds logN searchsteps to each bit adjustment procedure. For M CR cells,the MLWF adds a linear complexity with regard to the Mcoordinates to be searched. Therefore the total complexity of
the MLWF algorithm is O(MNlog2N). Table II comparesthe complexity of the MLWF algorithm with other resource
allocation algorithms. R is the total bits allocated to all SUs.From Table II, we can observe the complexity of the MLWF
algorithm is the lowest compared to the others.
We analyze the convergence of the MLWF briefly. In
theory, if we take an iterative method using the (13) and
(14), the algorithm converges slowly. In practical, the MLWF
0.001 0.01 0.1 1.0 2.0 4.02
4
6
8
10
12
14
Transmit power constraint of SUs (W)
Ave
rage B
its p
er
subca
rrie
r
Average Bits Per Subcarrier
MLWF
MultiCell MaxMin
Iterative Algorithm
Fig. 2. Average bits per subcarrier as a function of power limit
can approach to solution more closely after each adjustment.
Since the solution can be always improved to maintain the
interference constraints by reducing the mutual interference
in each adjustment procedure, the solution can be found after
certain number of adjustments.
IV. SIMULATION AND DISCUSSION
We conduct a series of experiments to assess the perfor-
mance of our proposed multicell multi-level waterfilling algo-
rithm. Consider an OFDM-based multicell CR system where a
primary cell with two PUs and three CR cells coexist. All users
are located within a 3 3 km areas. The primary base stationis in the middle of the area and the secondary base stations are
randomly distributed. Each PU or SU is uniformly distributed
within a 500-meter circle of its corresponding base station.The path loss exponent is 4, the variance of the shadowingeffect is 10 dB, and the multipath fading is assumed to beRayleigh [16]. There are 16 subchannels. The noise powerof each subchannel is set to 1013W. The frequency bandsoccupied by PUs are generated randomly with the maximum
number of OFDM subchannels is 2N3 . The transmission power
of a PU is equal to the number of OFDM subchannel within
the PUs band and the interference threshold of all PUs are
set to 5 1013 W. The results presented in this section isobtained from over 1000 Monte Carlo simulations.To evaluate the performance of the MLWF, we compare
the average bits per subchannel of the MLWF with other two
schemes: Multicell Max-Min [10] and iterative algorithm [12].
Fig.2 illustrates the average bits per subchannel as a function
of the transmit power limit. It can be seen that the capacity
of all algorithm increase as the transmit power limit increases.
The solution obtained by the MLWF is better than the others,
suggesting that cooperative power allocation can achieve more
capacity than the greedy-like algorithms.
Fig.3 shows the time complexity of the three algorithms
mentioned above. The elapsed time is counted by the inbuilt
function tic-toc in Matlab. We can see the complexity of
the MLWF is much lower than the other algorithms just
as the analysis in section III. Moreover, when the transmit
power constraint of the SU (PSU) is small, greedy power
allocation procedure (step 3 in algorithm) dominates the whole
elapsed time because it needs more time to find the water
727
-
5
0.001 0.01 0.1 1.0 2.0 4.0 10.010
4
103
102
101
100
101
Average Time Elapsed
Transmit power constraint of SUs (W)
Ave
rage T
ime E
lapse
d
MLWF
MultiCell MaxMin
Iterative Algorithm
Fig. 3. Average time elapsed as a function of power limit
0.001 0.01 0.1 1.0 2.0 4.0 10.00
0.5
1
1.5
2
2.5
3
3.5
4
Transmit power constraint of SUs (W)
Ave
rage N
um
ber
of A
dju
stm
ents
Average Number of Adjustments
Fig. 4. Average number of adjustments as a function of power limit
level while no power (bits) adjustment is required [see Fig.4].
However, when PSU is large enough, cooperative power (bits)
adjustment (step 5 in algorithm) need to be performed many
times to meet the interference-temperature constraint resulting
more time consumption [see Fig.4 and Fig.5(b)]. Naturally,
if the PSU is appropriate, there is no need to require the
adjustment procedure [see Fig.4 and Fig.5(a)] and greedy
power allocation procedure does less iterations. As analysis
above, it is reasonable to see the elapsed time of MLWF is
lower at the point of PSU= 0.1.
Finally, we investigate the convergence of the MLWF. As
discussed in section III, one of the main computational loads
of the MLWF lies in the number of the adjustment procedure.
From Fig.4 and Fig.5, we observe the average number of the
adjustment procedure lies in a narrow range [0, 4] and therandom instance total adjustment lies in [0,30]. Conservatively,
we conclude the MLWF method is effective and efficient
according to the complexity and convergence analysis in
section III.
V. CONCLUSION
In this paper we studied the resource allocation problem
in a multicell OFDM-based CR system. We try to maximize
the sum capacity of the system under transmission power and
interference constraints. An efficient and effective multilevel
waterfilling (MLWF) algorithm is developed by jointly con-
sidering the co-channel and mutual interference. The proposed
algorithm achieves better capacity performance with a lower
complexity, comparing to other existing algorithms. Besides,
0 10 20 30 40 50 60 70 80 90 1001
0.5
0
0.5
1
Nu
mb
er
of A
dju
stm
en
ts
Random instance (a)
PSU = 0.1
Times of Adjustments :PSU = 0.1
0 10 20 30 40 50 60 70 80 90 1000
5
10
15
20
Nu
mb
er
of A
dju
stm
en
ts
Random instance (b)
PSU= 10.0
Times of Adjustments : PSU = 10.0
Fig. 5. Adjustments as a function of power limit
our proposed MLWF algorithm always converges fast and
stably, which makes it promising for practical applications.
REFERENCES
[1] F. C. Commission et al., Facilitating opportunities for flexible, efficient,and reliable spectrum use employing cognitive radio technologies, Etdocket, vol. 3, pp. 03108, 2003.
[2] J. Mitola III and G. Maguire Jr, Cognitive radio: making software radiosmore personal, IEEE Personal Communications, vol. 6, no. 4, pp. 1318, 1999.
[3] S. Haykin, Cognitive radio: brain-empowered wireless communication-s, IEEE Journal on Selected Areas in Communications, vol. 23, no. 2,pp. 201220, 2005.
[4] D. abri, S. Mishra, D. Willkomm, R. Brodersen, and A. Wolisz, Acognitive radio approach for usage of virtual unlicensed spectrum, inproceeding of 14th IST Mobile and Wireless Communications Summit,2005.
[5] T. Weiss and F. Jondral, Spectrum pooling: an innovative strategyfor the enhancement of spectrum efficiency, IEEE CommunicationsMagazine, vol. 42, no. 3, pp. S814, 2004.
[6] Y. Zhang and C. Leung, Resource allocation in an OFDM-basedcognitive radio system, IEEE Transactions on Communications, vol. 57,no. 7, pp. 19281931, 2009.
[7] S. Wang, Efficient resource allocation algorithm for cognitive OFDMsystems, IEEE Communications Letters, vol. 14, no. 8, pp. 725727,2010.
[8] Y. Zhang and C. Leung, A distributed algorithm for resource allocationin OFDM cognitive radio systems, IEEE Transactions on VehicularTechnology, vol. 60, no. 2, pp. 546554, 2011.
[9] S. Gao, L. Qian, and D. Vaman, Distributed energy efficient spectrumaccess in cognitive radio wireless ad hoc networks, IEEE Transactionson Wireless Communications, vol. 8, no. 10, pp. 52025213, 2009.
[10] V. Reddy, Resource allocation for OFDM-based cognitive radio sys-tems, Ph.D. dissertation, 2011.
[11] W. Yu, G. Ginis, and J. Cioffi, Distributed multiuser power control fordigital subscriber lines, IEEE Journal on Selected Areas in Communi-cations, vol. 20, no. 5, pp. 11051115, 2002.
[12] F. Huang, S. Wang, and S. Du, Resource allocation in OFDM-basedmulti-cell cognitive radio systems, in proceeding of 2011 20th AnnualWireless and Optical Communications Conference (WOCC), 2011, pp.15.
[13] W. Yu and R. Lui, Dual methods for nonconvex spectrum optimiza-tion of multicarrier systems, IEEE Transactions on Communications,vol. 54, no. 7, pp. 13101322, 2006.
[14] J. Papandriopoulos and J. Evans, Low-complexity distributed algo-rithms for spectrum balancing in multi-user DSL networks, in IEEEInternational Conference on Communications, 2006. ICC06., vol. 7,2006, pp. 32703275.
[15] J. M. Cioffi. (2008) Ee379a course reader. stanford university. [Online].Available: http://www.stanford.edu/class/ee379a
[16] A. Goldsmith and S. Chua, Variable-rate variable-power MQAM forfading channels, IEEE Transactions on Communications, vol. 45,no. 10, pp. 12181230, 1997.
728